Effective multiplicative noise suppression for underwater wireless optical communication via a three-branch physics-informed network.

Underwater wireless optical communication (UWOC) can effectively provide high-bandwidth wireless connectivity for underwater scenarios; however, its performance is worsened by multiplicative fading and additive system noise. Traditional denoising methods for UWOC typically rely on handcrafted rules and empirical parameter selection, with limitations to the complex and variable underwater optical channel. Here, we propose a three-branch physics-informed network (TBPIN) to suppress the multiplicative noise of UWOC effectively. By embedding the UWOC channel equation into a composite loss function, the proposed network explicitly decouples the distorted received signal into three physically constrained components. This network is verified both on simulated and experimental UWOC datasets. The results show that TBPIN achieves advanced performance in signal-to-noise ratio gain and mean squared error reduction compared with traditional and other deep learning methods. This method provides an accurate and interpretable solution for multiplicative noise suppression in UWOC.